[martin-conur]

Summary

Every loadout's install.sh now builds the task-monitor TUI with cargo and links it into ~/.local/bin, so you no longer have to build it by hand. If cargo isn't installed it says so in one line and carries on, and a failed build only warns; the bash commands install either way.

Closes #277

What changed

• New lib/monitor-install.sh (aw_install_monitor), called from the drift-guarded install-shared-symlinks stanza, so all seven loadouts get it.
• It reads the binary path off cargo's JSON output (--message-format=json-render-diagnostics) instead of assuming monitor/target/release, so CARGO_TARGET_DIR / build.target-dir work. When that puts the binary outside the checkout, where it can't find bin/task-monitor-snapshot beside itself, the installer says to export TASK_MONITOR_SNAPSHOT.
• A reported executable that isn't there is treated as a failed build, never linked.
• Per the scope change on the issue, the CI Rust job is already in task-monitor 4/6: ratatui front end (repo view + all-repos overview, real-time): the task-monitor binary in monitor/ #282; this PR is the install part only.

Tests

• tests/install_symlinks.bats puts a stub cargo on PATH in setup, so no installer test runs a real release build. Five new tests: build args + link target, one-line skip with no cargo anywhere on PATH, failed build, missing reported executable, outside-the-checkout hint.
• Six mutations of the helper (no cargo check, no -x check, fail path returning 1, no hint, wrong link target, no --release) each turned at least one test red.
• Real end-to-end: ran claude-gh/install.sh against a scratch HOME with the real cargo; it built, linked and task-monitor --help ran through the link. Repeated with an outside CARGO_TARGET_DIR; the hint printed.
• ./run_tests.sh: 1466/1466 pass. tools/check-drift.sh clean, shellcheck -x clean on the changed files.

Docs

README task-monitor section, .claude/gh-workflow.md, the seven steering templates, CHANGELOG (with the upgrading note: re-run the loadout's install.sh; no task-init re-run needed).

🤖 Generated with Claude Code

[martin-conur]

Spec compliance

Issue #277 as amended by its scope-change comment (CI Rust job moved to #282; confirmed present in .github/workflows, with build/test/clippy/fmt on ubuntu and macos plus an MSRV job). Against the remaining scope:

• cargo build --release when cargo is on PATH, binary linked into ~/.local/bin: delivered via lib/monitor-install.sh, called from the drift-guarded install-shared-symlinks stanza in all 7 installers.
• One-line skip without cargo, bash commands still install: delivered and tested.
• install_symlinks tests: five new ones plus a cargo stub in setup, so no installer test does a real build.
• README, CHANGELOG with the upgrading note, workflow doc and the 7 steering templates: all updated. A grep finds no leftover "install.sh does not build it" wording.
• Prebuilt binaries are not touched, as specified.

Code-review findings

No correctness blockers.

• Under set -euo pipefail, binary=$(cargo … | sed … | tail -n 1) || binary="" is safe: a cargo failure yields an empty path and the warn-and-continue branch.
• Taking the last "executable" in cargo's JSON is right. Build scripts compile first and the crate has a single [[bin]].
• The -x check on the reported path stops a dangling symlink.
• The outside-the-checkout test uses pwd -P on both sides, so it is not fooled by a symlinked tmp dir such as macOS /var vs /private/var.
• CI ran on head 8af468e with all 7 checks green (bats ubuntu and macos, ShellCheck, drift, Rust ×3). That includes the Rust job from task-monitor 4/6: ratatui front end (repo view + all-repos overview, real-time): the task-monitor binary in monitor/ #282.

Nit, optional: the installer's cargo build has no --locked, while CI uses it. A stale Cargo.lock would be silently updated in the user's checkout. Adding --locked means changing the stub's expected args in the test as well. Not blocking.

PR body opens with ## Summary.

Verdict

clean-with-nits (the nit is optional). Merge decision is PM's.